École Nationale Supérieure de
Techniques Avancées

Abstract

Monitoring the condition and conformity of road markings is an essential road safety operation. A field analysis would involve closing the roads and is therefore very restrictive for the network manager. OFFROAD, which already produces 3D Lidar maps for various uses in the public works sector, would like to use its data to carry out this monitoring in an automated way, one of the steps in the process being the detection and classification of road markings. The aim of this internship is to set up the semantic segmentation deep learning protocol required for the automatic analysis of a road, and to extract from it the positions of the lines drawn and their type.